1 / 47

Example: Prime Generation

Example: Prime Generation. Problem: “in practice” efficiently generate all primes for a function Solution: Use recursive-evaluation method on “cofactors”. Cofactors. Partial evaluation of function F a = F evaluated at a=1 F a’ = F evaluated at a=0 Example: F=a’b’d’ + abd’ +a’b

paul2
Télécharger la présentation

Example: Prime Generation

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Example: Prime Generation • Problem: “in practice” efficiently generate all primes for a function • Solution: Use recursive-evaluation method on “cofactors” CS 150 - Spring 2008 – Lec #3: Combinational Logic - 1

  2. Cofactors • Partial evaluation of function • Fa = F evaluated at a=1 • Fa’ = F evaluated at a=0 • Example: F=a’b’d’ + abd’ +a’b • Fa = bd’ • Fa’ = b’d’ + b • For any function F, variable a: • F = a Fa + a’ Fa’ (“Shannon Expansion”) • Example: F = a bd’ + a’(b’d’ + b) [Check it!] CS 150 - Spring 2008 – Lec #3: Combinational Logic - 2

  3. Cofactors and Primes • Theorem: Any Function F, variable a, Primes are maximal implicants of set: • {a Primes(Fa), a’ Primes(Fa’), Primes(Fa) Primes(Fa’)} • In other words • Find the primes of Fa, multiply each by a • Find the primes of Fa’, multiply each by a’ • Multiply each prime of Fa by each prime of Fa’ • Maximal members of these sets are primes of F CS 150 - Spring 2008 – Lec #3: Combinational Logic - 3

  4. A X 0 X 1 0 0 1 1 D 0 0 0 0 1 1 0 X C B Example 0 0 1 1 Primes(Fa’) = d, bc Fa’ = 1 1 0 X F = Primes(F) = a’d, a’bc, abc’,ac’d’, bc’d A X 0 X 1 Primes(Fa) = bc’,c’d’ Fa = 0 0 0 0 CS 150 - Spring 2008 – Lec #3: Combinational Logic - 4

  5. Similar Fast Techniques to Generate Rows • Recurse down cofactor tree, throwing out primes when they disagree • Stop when: • Cofactor of on-set is empty (no rows) • All the remaining primes cover the whole space (1 row, with those primes) • Still one very fast technique, “Espresso-Signature” (to follow approximate algorithm) CS 150 - Spring 2008 – Lec #3: Combinational Logic - 5

  6. Approximate Algorithm • Don’t try to generate minimum cover • Given a cover (sum-of-products for on-set and don’t-care set) • Generate a prime cover smaller than given cover • Keep generating smaller prime covers until you can’t anymore • Program – Espresso-II CS 150 - Spring 2008 – Lec #3: Combinational Logic - 6

  7. Espresso-II algorithm (approx) • Given Cover C = c1 + c2 +…+cn • While (we keep making improvement) • PrimeCover = 0 • Foreach cube c_i of the cover, C • Find a prime p_i containing c_i that contains as many c_j’s as possible • PrimeCover = PrimeCover + p_i • We now have a prime cover p1+…+pn, Make the cover “irredundant” (example next slide). Result is a smaller prime cover • “Reduce” each cube away from a prime while still retaining a cover (this permits a different expansion in next iteration) CS 150 - Spring 2008 – Lec #3: Combinational Logic - 7

  8. Espresso-II example • F=a’b’d’ + abc’d’ + abc+ a’b, DC = abcd • Expand: • a’b’d’ => a’d’ • abc’d’ => bd’ • abc => bc • a’b => a’b • Prime Cover a’d’ + bd’ + bc + a’b • Prime bc is redundant: covered by bd’ + a’b + don’t-care • Irredundant cover is: a’d’ + bd’ + a’b • Reduce to a’b’d’ + bd’ + a’bd for next iteration CS 150 - Spring 2008 – Lec #3: Combinational Logic - 8

  9. Espresso-II presented here very simplified • See text for fuller description • What I left out • Details of EXPAND (how to pick the primes) • Details of IRREDUNDANT • Details of REDUCE • ESSENTIAL_PRIMES (identified early and tossed into don’t-care set for efficiency) • LAST_GASP (radical reduction to get new starting point) • MAKESPARSE (final pruning of literals in the output plane) CS 150 - Spring 2008 – Lec #3: Combinational Logic - 9

  10. Return to Exact Methods • Realization (mid-nineties): The only purpose of generating primes was to get to covering table! • Specifically: to get to “non-dominated” rows of covering table • Realization (beyond the scope of this class) • Can identify non-dominated rows without generating primes or minterms! • “Espresso Signature”: first radical revision of exact two-level synthesis algorithm CS 150 - Spring 2008 – Lec #3: Combinational Logic - 10

  11. Espresso-Signature • Espresso-Signature Algorithm • Generate non-dominated rows of covering table • This forms a cover, the “canonical cover” of a logic function • Use modified Espresso EXPAND to generate non-dominated columns of covering table • Solve the covering problem • Much more efficient than standard algorithm CS 150 - Spring 2008 – Lec #3: Combinational Logic - 11

  12. Two-Level Synthesis Summary • Two exact methods: Quine-McCluskey and Espresso-Signature • In practice, both highly efficient • Theoretically: exponential complexity • Algorithms work well because they fit circuits people actually design (though this set is not characterized well) • One approximate method shown, Espresso • Others (McBoole, e.g.) similar • Also theoretically exponential complexity! • In practice, fast, gives results within a few percent of optimum. CS 150 - Spring 2008 – Lec #3: Combinational Logic - 12

  13. Implementations of Two-level Logic • Sum-of-products • AND gates to form product terms(minterms) • OR gate to form sum • Product-of-sums • OR gates to form sum terms(maxterms) • AND gates to form product CS 150 - Spring 2008 – Lec #3: Combinational Logic - 13

  14. Two-level Logic Using NAND Gates • Replace minterm AND gates with NAND gates • Place compensating inversion at inputs of OR gate CS 150 - Spring 2008 – Lec #3: Combinational Logic - 14

  15. Two-level Logic Using NAND Gates (cont’d) • OR gate with inverted inputs is a NAND gate • de Morgan's: A' + B' = (A • B)' • Two-level NAND-NAND network • Inverted inputs are not counted • In a typical circuit, inversion is done once and signal distributed CS 150 - Spring 2008 – Lec #3: Combinational Logic - 15

  16. Two-level Logic Using NOR Gates • Replace maxterm OR gates with NOR gates • Place compensating inversion at inputs of AND gate CS 150 - Spring 2008 – Lec #3: Combinational Logic - 16

  17. Two-level Logic Using NOR Gates (cont’d) • AND gate with inverted inputs is a NOR gate • de Morgan's: A' • B' = (A + B)' • Two-level NOR-NOR network • Inverted inputs are not counted • In a typical circuit, inversion is done once and signal distributed CS 150 - Spring 2008 – Lec #3: Combinational Logic - 17

  18. Combinational Logic Summary • Logic functions, truth tables, and switches • NOT, AND, OR, NAND, NOR, XOR, . . ., minimal set • Axioms and theorems of Boolean algebra • Proofs by re-writing and perfect induction • Gate logic • Networks of Boolean functions and their time behavior • Canonical forms • Two-level and incompletely specified functions • Simplification • Two-level simplification CS 150 - Spring 2008 – Lec #3: Combinational Logic - 23

  19. Hardware Description Languages: Verilog • Verilog • Structural Models • (Combinational) Behavioral Models • Syntax • Examples CS 150 - Spring 2008 – Lec #3: Combinational Logic - 24

  20. Quick History of HDLs • ISP (circa 1977) - research project at CMU • Simulation, but no synthesis • Abel (circa 1983) - developed by Data-I/O • Targeted to programmable logic devices • Not good for much more than state machines • Verilog (circa 1985) - developed by Gateway (now Cadence) • Similar to Pascal and C • Delays is only interaction with simulator • Fairly efficient and easy to write • IEEE standard • VHDL (circa 1987) - DoD sponsored standard • Similar to Ada (emphasis on re-use and maintainability) • Simulation semantics visible • Very general but verbose • IEEE standard • “Synchronous/Reactive Languages” (1980’s-1990’s) • Esterel, Lustre, Signal, SMV, V++ • Developed for embedded software and formal verification • V++ Explicit “Hardwware Design Language” • Semantics much closer to hardware (but no tool support ) CS 150 - Spring 2008 – Lec #3: Combinational Logic - 25

  21. Design Methodology HDLSpecification Structure and Function(Behavior) of a Design Simulation Synthesis Verification: Design Behave as Required? Functional: I/O Behavior Register-Level (Architectural) Logic-Level (Gates) Transistor-Level (Electrical) Timing: Waveform Behavior Generation: Map Specification to Implementation CS 150 - Spring 2008 – Lec #3: Combinational Logic - 26

  22. Verilog/VHDL • The “standard” languages • Very similar • Many tools provide front-ends to both • Verilog is simpler • Less syntax, fewer constructs • VHDL supports large, complex systems • Better support for modularization • More grungy details • “Hello world” is much bigger in VHDL CS 150 - Spring 2008 – Lec #3: Combinational Logic - 27

  23. Verilog • Supports structural and behavioral descriptions • Structural • Explicit structure of the circuit • How a module is composed as an interconnection of more primitive modules/components • E.g., each logic gate instantiated and connected to others • Behavioral • Program describes input/output behavior of circuit • Many structural implementations could have same behavior • E.g., different implementations of one Boolean function CS 150 - Spring 2008 – Lec #3: Combinational Logic - 28

  24. Key Problem (Both Verilog AND VHDL) • Original use as modeling and simulation tools • Goal was to let designers simulate circuits • Synthesis came later • Semantics defined in terms of discrete-event simulation, not hardware • Leads to peculiar behavior, particularly concerning latches • Latches don’t necessarily appear when you think they should (“reg” declaration doesn’t necessarily mean a latch) • Latches appear when you aren’t expecting them (when Verilog decides there is undefined state) • US Patent 6,725,187: “Compiling Verilog into Timed Finite State Machines” • Oy, vey… • Key lesson: be careful! CS 150 - Spring 2008 – Lec #3: Combinational Logic - 29

  25. Verilog Introduction • the module describes a component in the circuit • Two ways to describe: • Structural Verilog • List of components and how they are connected • Just like schematics, but using text • Hard to write, hard to decode • Useful if you don’t have integrated design tools • Behavioral Verilog • Describe what a component does, not how it does it • Synthesized into a circuit that has this behavior CS 150 - Spring 2008 – Lec #3: Combinational Logic - 30

  26. Structural Model • Composition of primitive gates to form more complex module • Note use of wire declaration! module xor_gate (out, a, b); input a, b; output out; wire abar, bbar, t1, t2; inverter invA (abar, a); inverter invB (bbar, b); and_gate and1 (t1, a, bbar); and_gate and2 (t2, b, abar); or_gate or1 (out, t1, t2); endmodule By default, identifiersare wires CS 150 - Spring 2008 – Lec #3: Combinational Logic - 31

  27. Structural Model • Example of full-adder module full_addr (A, B, Cin, S, Cout); input A, B, Cin; output S, Cout; assign {Cout, S} = A + B + Cin; endmodule module adder4 (A, B, Cin, S, Cout); input [3:0] A, B; input Cin; output [3:0] S; output Cout; wire C1, C2, C3; full_addr fa0 (A[0], B[0], Cin, S[0], C1); full_addr fa1 (A[1], B[1], C1, S[1], C2); full_addr fa2 (A[2], B[2], C2, S[2], C3); full_addr fa3 (A[3], B[3], C3, S[3], Cout); endmodule Behavior Structural CS 150 - Spring 2008 – Lec #3: Combinational Logic - 32

  28. Simple Behavioral Model • Combinational logic • Describe output as a function of inputs • Note use of assign keyword: continuous assignment module and_gate (out, in1, in2); input in1, in2; output out; assign out = in1 & in2; endmodule Output port of a primitive mustbe first in the list of ports Restriction does not apply tomodules CS 150 - Spring 2008 – Lec #3: Combinational Logic - 33

  29. Verilog Module • Corresponds to a circuit component • “Parameter list” is the list of external connections, aka “ports” • Ports are declared “input”, “output” or “inout” • inout ports used on tri-state buses • Port declarations imply that the variables are wires module name ports module full_addr (A, B, Cin, S, Cout);input A, B, Cin;output S, Cout;assign {Cout, S} = A + B + Cin;endmodule inputs/outputs CS 150 - Spring 2008 – Lec #3: Combinational Logic - 38

  30. Verilog Continuous Assignment • Assignment is continuously evaluated • assign corresponds to a connection or a simple component with the described function • Target is NEVER a reg variable use of Boolean operators(~ for bit-wise, ! for logical negation) assign A = X | (Y & ~Z); assign B[3:0] = 4'b01XX; assign C[15:0] = 16'h00ff; assign #3 {Cout, S[3:0]} = A[3:0] + B[3:0] + Cin; bits can take on four values(0, 1, X, Z) variables can be n-bits wide(MSB:LSB) use of arithmetic operator multiple assignment (concatenation) delay of performing computation, only used by simulator, not synthesis CS 150 - Spring 2008 – Lec #3: Combinational Logic - 39

  31. Comparator Example module Compare1 (A, B, Equal, Alarger, Blarger); input A, B; output Equal, Alarger, Blarger; assign Equal = (A & B) | (~A & ~B); assign Alarger = (A & ~B); assign Blarger = (~A & B);endmodule CS 150 - Spring 2008 – Lec #3: Combinational Logic - 40

  32. Comparator Example // Make a 4-bit comparator from 4 x 1-bit comparatorsmodule Compare4(A4, B4, Equal, Alarger, Blarger); input [3:0] A4, B4; output Equal, Alarger, Blarger; wire e0, e1, e2, e3, Al0, Al1, Al2, Al3, B10, Bl1, Bl2, Bl3; Compare1 cp0(A4[0], B4[0], e0, Al0, Bl0); Compare1 cp1(A4[1], B4[1], e1, Al1, Bl1); Compare1 cp2(A4[2], B4[2], e2, Al2, Bl2); Compare1 cp3(A4[3], B4[3], e3, Al3, Bl3); assign Equal = (e0 & e1 & e2 & e3); assign Alarger = (Al3 | (Al2 & e3) | (Al1 & e3 & e2) | (Al0 & e3 & e2 & e1)); assign Blarger = (~Alarger & ~Equal);endmodule CS 150 - Spring 2008 – Lec #3: Combinational Logic - 41

  33. Simple Behavioral Model: the always block • always block • Always waiting for a change to a trigger signal • Then executes the body module and_gate (out, in1, in2); input in1, in2; output out; reg out; always @(in1 or in2) begin out = in1 & in2; end endmodule Not a real register!! A Verilog register Needed because of assignment in always block Specifies when block is executed I.e., triggered by which signals CS 150 - Spring 2008 – Lec #3: Combinational Logic - 42

  34. always Block • Procedure that describes the function of a circuit • Can contain many statements including if, for, while, case • Statements in the always block are executed sequentially • (Non-blocking assignments <= are executed in parallel) • Blocking assignments = are executed sequentially • Entire block is executed at once • Final result describes the function of the circuit for current set of inputs • intermediate assignments don’t matter, only the final result • begin/end used to group statements CS 150 - Spring 2008 – Lec #3: Combinational Logic - 43

  35. “Complete” Assignments • If an always block executes, and a variable is not assigned • Variable keeps its old value (think implicit state!) • NOT combinational logic  latch is inserted (implied memory) This is usually not what you want: dangerous for the novice! • Any variable assigned in an always block should be assigned for any (and every!) execution of the block • Check particularly: case statements (missing default, and how can the compiler tell?) CS 150 - Spring 2008 – Lec #3: Combinational Logic - 44

  36. Incomplete Triggers • Leaving out an input trigger usually results in a sequential circuit module and_gate (out, in1, in2); input in1, in2; output out; reg out; always @(in1 or in2) begin out = in1 & in2; end endmodule module and_gate (out, in1, in2); input in1, in2; output out; reg out; always @(in1) begin out = in1 & in2; end endmodule Generates a Latch Doesn’t Generate a Latch CS 150 - Spring 2008 – Lec #3: Combinational Logic - 45

  37. One Latch or Two? module d_latch (clk, a, c); input clk, a; output c; reg c; reg b; always @(posedge clk) begin b = a; c = b; end endmodule CS 150 - Spring 2008 – Lec #3: Combinational Logic - 46

  38. One Latch or Two? module d_latch (clk, a, c); input clk, a; output c; reg c; reg b; always @(posedge clk) begin b <= a; c <= b; end endmodule • Why? • Non-blocking assignments in parallel • c gets previous value of b CS 150 - Spring 2008 – Lec #3: Combinational Logic - 47

  39. Verilog if • Same as C if statement // Simple 4:1 mux module mux4 (sel, A, B, C, D, Y); input [1:0] sel; // 2-bit control signal input A, B, C, D; output Y; reg Y; // target of assignment always @(sel or A or B or C or D) if (sel == 2’b00) Y = A; else if (sel == 2’b01) Y = B; else if (sel == 2’b10) Y = C; else if (sel == 2’b11) Y = D; endmodule CS 150 - Spring 2008 – Lec #3: Combinational Logic - 48

  40. Verilog if • Another way // Simple 4:1 mux module mux4 (sel, A, B, C, D, Y); input [1:0] sel; // 2-bit control signal input A, B, C, D; output Y; reg Y; // target of assignment always @(sel or A or B or C or D) if (sel[0] == 0) if (sel[1] == 0) Y = A; else Y = B; else if (sel[1] == 0) Y = C; else Y = D; endmodule CS 150 - Spring 2008 – Lec #3: Combinational Logic - 49

  41. Verilog case • Sequential execution of cases • Only first case that matches is executed (implicit break) • Default case can be used // Simple 4-1 mux module mux4 (sel, A, B, C, D, Y); input [1:0] sel; // 2-bit control signal input A, B, C, D; output Y; reg Y; // target of assignment always @(sel or A or B or C or D) case (sel) 2’b00: Y = A; 2’b01: Y = B; 2’b10: Y = C; 2’b11: Y = D; endcase endmodule Conditions tested intop to bottom order CS 150 - Spring 2008 – Lec #3: Combinational Logic - 50

  42. Verilog case • Note: case (case_expression) case_item1 : case_item_statement1; case_item2 : case_item_statement2; case_item3 : case_item_statement3; case_item4 : case_item_statement4; default : case_item_statement5; endcase • … is the same as … if(case_expression == case_item1) case_item_statement1; else if (case_expression == case_item2) case_item_statement2; else if (case_expression == case_item3) case_item_statement3; else if (case_expression == case_item4) case_item_statement4; else case_item_statement5; CS 150 - Spring 2008 – Lec #3: Combinational Logic - 51

  43. Verilog case vs C switch • Note difference from C: switch (case_expression) { case_item1 : case_item_statement1; case_item2 : case_item_statement2; case_item3 : case_item_statement3; case_item4 : case_item_statement4; default : case_item_statement5; } • … is the same as … if(case_expression == case_item1) case_item_statement1; if (case_expression == case_item2) case_item_statement2; if (case_expression == case_item3) case_item_statement3; if (case_expression == case_item4) case_item_statement4; case_item_statement5; This is why we need break statements in C! CS 150 - Spring 2008 – Lec #3: Combinational Logic - 52

  44. Verilog case • Without the default case, this example would create a latch for Y • Assigning X to a variable means synthesis is free to assign any value // Simple binary encoder (input is 1-hot) module encode (A, Y); input [7:0] A; // 8-bit input vector output [2:0] Y; // 3-bit encoded output reg [2:0] Y; // target of assignment always @(A) case (A) 8’b00000001: Y = 0; 8’b00000010: Y = 1; 8’b00000100: Y = 2; 8’b00001000: Y = 3; 8’b00010000: Y = 4; 8’b00100000: Y = 5; 8’b01000000: Y = 6; 8’b10000000: Y = 7; default: Y = 3’bXXX; // Don’t care when input is not 1-hot endcase endmodule CS 150 - Spring 2008 – Lec #3: Combinational Logic - 53

  45. Verilog case (cont) • Cases are executed sequentially • Following implements a priority encoder // Priority encoder module encode (A, Y); input [7:0] A; // 8-bit input vector output [2:0] Y; // 3-bit encoded output reg [2:0] Y; // target of assignment always @(A) case (1’b1) A[0]: Y = 0; A[1]: Y = 1; A[2]: Y = 2; A[3]: Y = 3; A[4]: Y = 4; A[5]: Y = 5; A[6]: Y = 6; A[7]: Y = 7; default: Y = 3’bXXX;// Don’t care when input is all 0’s endcase endmodule CS 150 - Spring 2008 – Lec #3: Combinational Logic - 54

  46. Parallel Case • A priority encoder is more expensive than a simple encoder • If we know the input is 1-hot, we can tell the synthesis tools • “parallel-case” pragma says the order of cases does not matter // simple encoder module encode (A, Y); input [7:0] A; // 8-bit input vector output [2:0] Y; // 3-bit encoded output reg [2:0] Y; // target of assignment always @(A) case (1’b1) // synthesis parallel-case A[0]: Y = 0; A[1]: Y = 1; A[2]: Y = 2; A[3]: Y = 3; A[4]: Y = 4; A[5]: Y = 5; A[6]: Y = 6; A[7]: Y = 7; default: Y = 3’bX; // Don’t care when input is all 0’s endcase endmodule CS 150 - Spring 2008 – Lec #3: Combinational Logic - 55

  47. casex Example // Priority encoder module encode (A, valid, Y); input [7:0] A; // 8-bit input vector output [2:0] Y; // 3-bit encoded output output valid; // Asserted when an input is not all 0’s reg [2:0] Y; // target of assignment reg valid; always @(A) begin valid = 1; casex (A) 8’bXXXXXXX1: Y = 0; 8’bXXXXXX10: Y = 1; 8’bXXXXX100: Y = 2; 8’bXXXX1000: Y = 3; 8’bXXX10000: Y = 4; 8’bXX100000: Y = 5; 8’bX1000000: Y = 6; 8’b10000000: Y = 7; default: begin valid = 0; Y = 3’bX; // Don’t care when input is all 0’s end endcase end endmodule CS 150 - Spring 2008 – Lec #3: Combinational Logic - 57

More Related